For a Data-Driven Interpretation of Rules wrt GMP Conclusions in Abductive Problems

Adrien Revault d'Allonnes (1) , Herman Akdag (1) , Bernadette Bouchon-Meunier (1)

1 MALIRE - Machine Learning and Information Retrieval

Abductive reasoning is an explanatory process in which potential causes of an observation are unearthed. In its classical – crisp – version it offers little lattitude for discovery of new knowledge. Placed in a fuzzy context, abduction can explain observations which did not, originally, exactly match the expected conclusions. Studying the effects of slight modifications through the use of linguistic modifiers was, therefore , of interest in order to describe the extent to which observations can be modified yet still explained and, possibly, create new knowledge. We will concentrate on the formal definition of fuzzy abduction given by Mellouli and Bouchon-Meunier. Our results will be shown to be incompatible with established theories. We will show where this incompatibility comes from and derive from it a selection of fuzzy implication , based on observable data.

Mots clés: Abductive reasoning; fuzzy inference; fuzzy implications; Generalised Modus Ponens

Domaines: Intelligence artificielle [cs.AI]

Adrien Revault d'Allonnes, Herman Akdag, Bernadette Bouchon-Meunier. For a Data-Driven Interpretation of Rules wrt GMP Conclusions in Abductive Problems. Journal of Uncertain Systems, 2009, 3 (4), pp.280-297. ⟨hal-00600708⟩